Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Chapter 5 SOME SPECIFIC MATERIALS In this chapter, several specific materials or classes are described. These examples will give a flavor of the complex compromises that must be made in optimizing material and process choices . Even so, the presentation is simplified, and it should be emphasized that materials engineering for packaging and interconnection follows requirements set by system design considerations. Thus, at any given time, system needs may put pressure on a specific property. Cost, in general, is a critical issue, although it is not susceptible to treatment in a report of this type. Cost is both volume- and process-dependent, further complicating the material design choice. THE EVOLUTION OF EPOXY MATERIALS IN PLASTIC PACKAGING \7arious epoxies have been developed in recent years, and today they play an important role in electronic packaging. The following sections cover some of the special features and how they can be modified to meet specific needs. Epoxy Versus Silicone Materials The earliest materials used for plastic packaging of microelectronic devices were silicones because of their high-temperature performance and high purity. The common bisphenol-A (BPA) epoxies were introduced that have lower glass transition temperatures (T~) . Novolac epoxies are generally preferred over BPA materials because of their higher functionality and attendant improvements in heat re s i s Lance Howeve r, epoxie s can have high ionic impurity levels because their reactior~ chemistry uses an excess of halogen- containing epichlorohydrin. Hybrid encapsulant materials consisting of epoxies and silicones captured a large segment of the molding compound market in the mid- 1970s because they combined some of the high- temperature performance and the high purity of the si li cones witch the mechanical properties and the solvers yes istanc~ of the epoxies . Molding compounds s ince then have moored steadily toward all - epoxy systems ~ as the high- temperature novolac materials were improved and the ionic impurity levels were driven below 100 ppm. Figure 5-1 charts the property and process evolution. 71
72 1 970 1 980 1 990 DRAM LOGIC (# GATES)(CMOS) 1 6K 64K 256K 1 M 4M 0.1K 1K 10K 100K l | FLAME RETARDED3 l | LOW IONIZ} BLE CHLORINEl | LOW STRESS I | LOW a- EMISSIONl , ANi1YDRIDE CURE f r I AMINE CURE NOVOLAC a. ~ STRESS MODIFICATION . rSYNTHETIC SiO2 flLLER THERMAL STRESS (MPa) THERMAL EXPANSION(x10~6/°K) IONIZABLE CHLORINE (ppm} UlTh CONTENT (ppb) FLAME RETARDANCY CURE TIME (SECONDS) MELT VISCOSITY (POISE) (MIN.) MOISTURE SENSITIVITY (REL.) TEMPERATURE CYCLING (REL.) 12 1 10 1 00 500 Figure 5-1. Property improvement history of epoxy encapsulating compounds, 1970-1990. Low-Stress Materials A major advance in epoxy molding compounds was the development of the low-stress materials in the early 1980s. Package cracking, passivation-layer cracking, and circuit-pattern deformation can all result from the disparities in the coefficients of thermal expansion among chip, plastic package, and metal leadframe. This problem was exacerbated as the surface area of memory devices with 64K and 256K DRAMs became a significant fraction of the package itself. The plastic edge thickness became thin, and cracking began to Occur. To alleviate this problem, the low-stress molding compounds combined several ,, ~ alstr1 outlons were used to increase the filler loading to over 72 percent by weight, thereby reducing the coefficient of thermal expansion by as much as 25 percent (21 x 10-5 to 16 x 10-5 in./in.°C). The second advance was the addition of elastomeric modifiers in a dispersed domain morphology, which improves toughness. Research on the size, size distribution, and interracial region was a contributor to the improved performance of these materials. It is at point that the Japanese suppliers surpassed the domestic suppliers in compound performar~ce. approaches. Fillers we to rent size and snare The s econd advance war the major this molding
73 Low-Alpha Materials Alpha particles emitted from the silica fillers can cause soft errors in memory devices. The development of molding compounds that were low-alpha- particle emitters was ~ major advancement. Although these materials were available from the late 1970s, the silica was from the few mines that had low natural uranium and thorium contamination. Supplies were scarce, and prices were much higher than conventional materials ($25/lb compared to $3.50/lb). The use of synthetic silica in the early 1980s reduced the price of low-alpha materials to $10/lb by 1986. These materials quickly captured the market for memory devices? effectively eliminating silicone rubber in applications where it was used for alpha-particle protection. Processing Improvements Processing improvements also were required to keep pace with the increasing demands of higher-pin-count packages. The 40-pin DIP of the early 1980s was the first design to experience significant flow-induced stress problems. The long cantilever of the leads and the close proximity of the wire bonds made the package much more susceptible to lead movement, paddle shift, and wire sweep; the forces are proportional to the product of the material viscosity and its velocity. Japanese suppliers led the way in developing materials that had significantly lower viscosities than the previous generation of materials. The introduction of fine-pitch packages and thin leadframes has continued to push the need for very-low-viscosity molding compounds. The viscosity at the molding temperature has been reduced from 1500 poise in 1980 to 300 poise (at 100 sect) in 1988, while both the filler loading and crack resistance of the material have been increased. It is also significant that the cure time for epoxy molding compounds has been reduced from 3 to 4 minutes in the mid- 1970s to 60 to 90 seconds in the late 1980s, thereby contributing an important productivity increase. Market Shares of the Maj or Molding Compound Suppliers Japanese firms now hold 70 percent of the world market share in epoxy molding compounds for IC packaging. Less than a decade ago, U. S . companies had the maj ority of the market. This substantial U. S . market share advantage waned with the introduction of newer low-stress epoxy novolac materials. High purity, low viscosity, and low shrinkage stress were the major technology areas where the domestic suppliers fell behind. FUTURE TRENDS IN PLASTIC PACKAGING MATERIALS Advancements are needed to meet the challenge of future high-density packaging requirements. Some of these are discussed in the following sections.
74 Surface-Mount Technology Surface-mount attachment of IC packages will grow very rapidly over the next five years. The use of vapor phase and IR reflow solder methods for surface-mount devices exposes the entire molded body to high temperatures that can cause cracking through two different mechanisms. The first is by the thermal shrinkage forces already discussed. Continued reduction in the coefficient of thermal expansion, reduction of the glassy modulus, and additional improvements in adhesion will be needed to prevent thermal- shrinkage stress cracking of the molded body. The second crack mechanism is through rapid vaporization of moisture within the molded body, which causes volume expansion above the glass transition temperature (T~) of the plastic package. To alleviate this serious problem, there are four possible material improvements that could be made: a) lower the moisture uptake of the molding compound, b) increase the Tg to near the solder reflow temperature, c) increase the mechanical modulus above Tg to minimize high- temperature deformation, and d) improve adhesion to the bottom surface of the paddle, where delamination occurs. Very High Pin Count Packages Pin counts for microprocessors and logic gate arrays will continue to rise rapidly over the next several years. This will cause package dimensions to grow significantly despite the anticipated reductions in lead spacing. This trend will continue to put pressure on the processability of epoxy molding compounds. Larger packages with thin lead frames and very narrow spacings between the leads and wire bonds are much more susceptible to flow- induced stresses. Continued decreases in molding compound viscosity will be needed to package the 196- and 244-pin packages that will appear in the next few years, since current materials and processes may not be able to realize high production yields on these designs. New process technology, such as the multiplunger transfer molding and smaller conventional molds, may be required to provide the long cavity fill times needed to reduce the flow-induced stresses, if further viscosity reductions cannot be achieved. Pin and pad grid arrays (PGAs) could show substantial volume growth if the lower productivity of molded high-pin-count packages increases their costs close to that of the PGAs ; the PGAs generally are easier to assemble onto printed wiring boards than fine-pitch chip carriers that are restricted to edge leads. Heat Dissipation Power dissipation requirements of IC devices have been steadily rising. Although the shift from ECL to CMOS technology has at present substantially reduced the problem, it is only a postponement. Power increases rapidly because power dissipation scales linearly with feature size at constant voltage, but on-chip circuit density increases as the square of the feature- size reduction ratio. Continued growth in device integration and miniaturization will push the power per chip to over 5 W in the near future, a level that cannot be easily dissipated in a conventional plastic package
75 design. Thermal management will become much more important than it is at present, and the epoxy molding compound will have to play a major role in this management if the costs of plastic packages are to remain low. High- conductivity fillers are available, such as alumina, silicon nitride, silicon carbide , and magnesium oxide ~ see Appendix E) . These alternative materials are all very abrasive. Developments in tool design may enable manufacturers to use these more abrasive materials, or surface treatments for the fillers could be developed to reduce their abrasiveness. The dielectric properties of these alternative fillers are not as good as silica (see Appendix E). In the absence of these technical developments, active thermal-management components will have to be added to the packaging. Devices such as molded-in heat spreaders or added-on cooling fins could significantly affect package manufacture and assembly, however, with attendant increases in cost. Figure 5-1 recounts the improvements in thermoses molding compounds since the 1970s, and the expected needs over the next few years. Plastic Alternatives to Epoxy Encapsulation Plastic encapsulation of integrated circuit chips is, as noted above, dominated by novolac-based epoxies that are heavily filled with silica powder. These compounds are ther~osetting (i.e., the resulting polymer is highly cross-linked) and the transfer molding cycle times for processing are long compared winch injection-molding cycle times . On the other hand, transfer molds typically have hundreds of cavities producing hundreds of parts per cycle. The viscosity of a thermosetting epoxy evolves in a complicated manner, and thermoplastic compounds could well prove eas ier to process . However, procedures for processing epoxies are widely practiced, and the technology is entrenched. Reuse of scrap, as an economy, is another reason for considering thermoplastic encapsulants. Liquid crystalline polymers form another class of material that is under cons iteration as a replacement for epoxy encapsulants. These materials are usually aromatic polyesters that contain naphthalene groups in the main chain. They form anisotropic liquids and solutions, and molded parts retain the anisotropic properties, an aspect that may be turned to advantage in a well- designed process because excellent mechanical properties can be achieved. This area is worthy of further research and engineering support to investigate the potential of liquid crystalline polymers in electronic packaging. Polyphenylene sulfide is a thermoplastic material that has been available for several years as an alternative to epoxies. No widespread use has been reported, but development continues. This material can produce corrosive acid by-products and must be compounded carefully to avoid problems that could arise. As with epoxies, polyphenylene sulfide is filled with an inorganic powder, usually silica, to reduce the TOE.
76 ORGANIC PRINTED CIRCUIT BOARD MATERIALS Printed circuit substrates for electronic systems are usually organic, although circuits in the high-performance area tend to have more ceramic- based representation. It is obvious from Figure 1- 8, discussed in Chapter 1, that chip technology in terms of density has advanced much faster than circuit board technology, conductor width and other features on ICs were 10 times smaller than those on circuit boards in 1965, and today they are about 100 times smaller. This linear gain is even more important in terms of circuit area. Progress in interconnection has not been as great. The large mismatch in dimensions that has now developed, less than 1 Am on the chip to more than 100 ,um circuit boards, will over the next few years almost certainly lead deco more extensive use of circuits of intermediate density. The dominant material for circui t boards for more than two decades is a glass - fabric - reinforced epoxy res in labeled FR-4 . This material is fire - retardant, stable, and amenable to high-volume processing. Inner layers in MLBs are available as B-stage material that can be processed to generate circuit traces and vies, then "layered up" and pressure-cured into the final, coordinated multilayer structure. This system has served the industry well and is likely to continue in widespread use for circuits that do not contain features smaller than about 100 Am (4 mils ~ . Advanced circuits appear to be moving to an improved version of FR-4 in which the bisphenol-A epoxy is replaced by a more highly crosslinked resin (novolac), a resin related to the molding compound employed for chip encapsulation. The glass reinforcement may also be modified or replaced to meet circuit demands. Thus, with only minor modification, it can be predicted that FR-4 will provide excellent circuits with features of about 20 Am in size. This development should provide MLBs of considerable sophistication for the remainder of this century. The availability of this improved FR-4 will blur the boundaries between very-high-performance circuit boards and multichip modules. Other organic materials have coexisted with FR-4. FR-2 is a paper- reinforced phenolic resin, an inexpensive substrate material for single- and double-sided circuits. This substrate is popular in Japan. Molded printed circuits have been manufactured mainly from polyether sulfone and polyether imide. Several forms of flexible circuitry exist, including polyimide film, polyester film, and glass-mat--reinforced polyester. Japanese consumer products employ polyimide film circuits. The glass-mat polyester also was used in telecommunications equipment. These materials (FR-2, flexible substrates, and some other examples) were (or are) niche products that are not part of the high-density substrate evolution history. Beyond the improved FR-4, there are a number of alternative substrate and dielectric polymers available to meet specific needs. Polyimides provide excellent high- temperature performances, and a great deal of engineering has been conducted on this class of materials. Bismaleimide-triazene polymers (BT resins) exhibit good high-temperature properties and are compatible with epoxies. Polytetrafluoroethylene and certain highly crosslinked hydrocarbons offer a low dielectric constant as well as high-temperature performance. As
77 the need arises for employing these more advanced materials, sources with the requisite expertise are available in the United States as well as in Japan. It remains to be seen which country will take the lead in producing materials and processes in this area. PROCESSING TECHNOLOGY FOR CERAMIC PACKAGES, BOARDS, AND SUBSTRATES The earliest packages for electronic circuitry were metal packages, and the interconnection between devices was generally done with wires. The earliest use of ceramics came in the form of bases for electron tubes with glass-sealed electrical feed-throughs. With the demand for miniaturization, some circuit designs began to use ceramic substrates. These provided a strong board upon which various wiring patterns could be placed. During the 1950s, the technology was developed for interconnecting various discrete devices using metal lines on these substrates. Furthermore, the technology for placing some resistors on these substrates in the form of thick films was also introduced. One of the first applications of the ceramic package came with the need to house a simple quartz oscillator. This also was done in the late 1950s. While the substrate technology had led to miniaturization of resistors and capacitors, the individual transistor devices were still packaged in metal cans with the typical 3 lead wires protruding from the bottom of the can. This design was inconsistent with a low profile, so the model of the package used for the quartz oscillator acted as the driver for the first ceramic bans is tor package . With the use of a ceramic substrate technology for interconnections, it quickly became obvious that, with increased complexity of circuitry, it was necessary to have a means of permitting one conductor path to cross another. A number of schemes were developed to do this, using complex insulated crossovers and small bridges, but the real technology for handling this problem came in two forms. These two forms, the ceramic multilayer tape process and the screen-printing process, constitute the basis for much of today's processing technology for ceramic packaging. Discussed in the following sections are some of the earliest technologies for making ceramic parts as well as the two major current processing technologies mentioned above. Dry Pressing The earliest process for making ceramic parts for electronic circuitry was that of dry pressing, and this included the substrates used for interconnection. In this technology, the powder is usually prepared by mixing the raw materials as a slurry in a ball mill. The material is dried and then calcined for phase formation and intial chemical reaction. The calcined mixture is again ball-milled to grind it into very fine particles, and this ground powder is then dried to prepare it for pressing. In the pressing operation, the powder is typically poured into a cavity of a metal die, and then the die plunger presses the powder into the desired shape. This
technology can be cumbersome and relatively expensive, does not lend itself to having any buried conductor paths, and gives a part with a limited surface smoothness. The Ceramic Tape Process The ceramic tape- casting process Sometimes known as the doctor-blade process ~ was adapted from a process typically used in fabricating sheets of polymers or organics. In the ceramic adaptation, ache powder is mixed with a liquid solvent containing various organic additives, which act as binders, dispersants, and plasticizers, to give a slurry with a viscosity roughly that of heavy cream. This slurry is poured into a reservoir and then spread into a thin sheet by having the slurry pass under the gate, known as the doctor blade, as either the reservoir is moved over a smooth sheet, or a smooth sheet is moved under the reservoir. After casting, the slurry is allowed to dry by volatilization of the solvent. The dried ceramic still contains sufficient organic binders so that the tape can be stripped from the carrier, and it exhibits sufficient flexibility and strength to be handled, cut, or punched. At this stage, the tape is referred to as a "green" tape, meaning an unsintered ceramic tape. The thickness is generally in the range of 0.5 to 1 mm . This ''green" tape can be used to fabricate substrates by punching the substrate shape from the rolled green tape, heating the piece to temperatures of about 200 to 300°C to pyrolyze or volatilize the organic binders, and finally heating to a sufficiently high temperature to sinter the ceramic particles into a dense, hard sheet. Because the particles are well dispersed in the original slurry and because the surface of the tape is very smooth, the sintered ceramic offers a high-quality, dense, pore-free, strong surface on which to place electronic interconnect devices. By about 1960, the concept of making a multilayer substrate or ceramic package was introduced. In this technology, some or all of the metal interconnect lines are buried within the final ceramic. This is accomplished by printing the metal lines on each of several sheets of the unsintered ceramic prior to fabricating the package. The metal paste is formed by mixing the metal particles with organic binders and then patterning the metal lines by forcing the paste through a screen covered by an emulsion into which the appropriate pattern had been photolithographically etched. These layers of interconnect traces, or conduction paths, can be connected from one layer to the next using through-holes known as "vias." These vies consist of holes punched in the green tape into which the metal ink is forced in the screen- printing operation, thereby providing a conduction path from a point in one layer a to similar point in the layer above or below it. After all the layers are printed, they are then stacked together, with the appropriate registration, and laminated under slight heat and pressure into a solid unsintered block of ceramic and metal. This laminated piece can then be heated to remove the organics and finally to sinter it into a dense, hard ceramic. Furthermore, by appropriately shaping some of the layers, a cavity can be formed in the package into which ultimately the active device, such as a silicon chip, can be placed.
79 Screen Printing Screen printing mentioned above, provides an alternative mechanism for producing a three-dimensional interconnect technology on a ceramic base. This technology was developed in the late 1950s, at about the same time as the multilayer ceramic tape technology, and provides some of the same functions. In a typical application, a ceramic substrate is used onto which various layers are screen-printed. The first of these layers might be an insulating layer that is screen-printed, generally over a broad area. After printing, the layer is typically dried and fired at some temperature lower than that at which the original substrate was fired. If an insulating layer thicker than that typically obtained by a single screen-printing layer is necessary, a subsequent layer may also be screen-printed on, dried, and fired again. A metallization layer may then be screen-printed through a screen with an appropriate pattern. This, too, is dried and fired. The next insulating layer can then be applied, typically allowing openings for vies between the metall~zation in the first layer and the second layer. These various print, dry, and fire operations can continue to make a multilayer interconnect board. In contrast to the multilayer green tape technology, this method does not easily allow for a cavity into which an active device can be sealed and protected in an hermetic environment. OTHER CERAMIC MATERIALS This section describes some other ceramic packaging materials and makes projections about those materials that could play a prominent role in the future. Glass and Porcelain The earliest use of electronic ceramics was as insulators for carrying telegraph lines, telephone lines, and power distribution lines. Ceramics and glasses also were used in the fabrication of electron tubes, both as a tube housing in the case of glass, and, in some tubes, as the base material with glass-to-metal feed-throughs for the electrical conductors. The earliest attempts at making ceramic substrates and ceramic packages employed the materials available at the time, which were typically glasses and porcelains. Porcelain is generally made of naturally-occurring materials, such as clays and talcs. These substrates suffered from low strength, poor thermal conductivity, and poor surface finish. Early in the use of ceramics for substrates and packages, aluminum oxide (alumina) became the major material of choice, and it continues so today. Typical compositions are made up of about 92 to 94 percent aluminum oxide, the balance being materials such as magnesium oxide (magnesia) or silicon dioxide (silica). The earliest metallization used was a molybdenum-manganese composition that had been known to fire onto aluminum oxide with a very adherent bond. This, too, was quickly replaced by either pure molybdenum or pure tungsten as the metallization of choice in co- fired aluminum oxide ceramic packages. These metallizat.ions continue to be the choice today. Both of these materials
80 can be made to sinter at about the same temperature as the aluminum oxide (about 1600°C) and provide reasonable electrical conductivity. However, both of these metal systems oxidize easily and require that the aluminum oxide be Wintered in a hydrogen or hydrogen-water vapor atmosphere. These alumina-based compositions provide a strong material with fairly high thermal conductivity. With tape processing, they can easily provide high-reliability hermetic packages for active semiconductor devices. Low-Fire Materials Low-firing ceramic materials are generally those that fire at temperatures below 1000°C, in contrast to the 1600 °C necessary for firing aluminum oxide. Although a few single-phase ceramic materials do sinter at temperatures below 1000°C, this not the general case. Consequently, typical low-fire materials are made from glass-ceramic composites that fall into two categories. The first is known as glass-ceramics or devitrifying glasses. A glass is first formed by melting of the constituent materials, and then the cooled glass is crushed and ground into a fine powder, known as a frit. The frit is formed into the desired shape using any of the fabrication technologies discussed earlier. The article is sintered into a dense, glassy material, which is then annealed at some lower temperature where the glass crystallizes (devitrifies) into a two-phase crystalline-glass composite. The second technology is known as glass-bonded ceramics. In this case, the desired crystalline phase is mixed as a powder with a frit of the desired glass phase. These again are fabricated into shapes by the various forming techniques already described and subsequently heated to sinter them into a dense material where the glass acts as a low-temperature sintering aid that bonds the ceramic particles together. The glass-bonded ceramic methodology is by far the most widely used. Aluminum oxide is again the typical crystalline material of choice. Its earliest use was in the screen-printed multilayer process described earlier. Here, the mixture of aluminum oxide and fritted glass is screen-printed onto a substrate and fired at temperatures typically between 800 and 1000 °C to make a dense ceramic layer. As the complexity of thick-film interconnect substrates increased, the number of print-dry-fire operations exceeded 40. To simplify the process and yet use the firing equipment already in place, some producers have recently switched to a low-firing ceramic tape technology. Naturally, the tape generated to fill this void is again a glass-bonded ceramic composite. Here, the aluminum oxide powder and fritted-glass mixture is tape cast as described earlier. These tape-cast layers can also be screen-printed with the various interlayer metallization patterns and vies for three-dimensional interconnect applications. This low-fire tape process offers at least two advantages. First, it permits use of the low- cost firing equipment available for thick- film printed circuitry. Second, it allows the use of metal conductors other than tungsten or molybdenum. Typical metals are silver, silver-palladium, or gold alloys, all of which have lower resistivity than molybdenum or tungsten and thus
81 contribute to faster circuit-speed designs. A current challenge in this technology is the possible use of copper metal as a low-cost, low-resistivity interconnect material. Implementation of this ideal is hampered by the difficulty of preventing oxidation of the copper in typical firing atmospheres. The usual glass phases, in the glass-bonded ceramic compositions, are lead silicate-based glasses. If these compositions are fired in the reducing or neutral atmosphere conditions necessary for maintaining copper in the metallic state, the lead will partially or fully reduce, leaving a lossy (low-resistivity) dielectric mater' al. Thus, the full implementation of copper metallizations awaits a technology that can successfully implement glass-bonding agents that are not easily reduced in such atmospheres. Laboratory successes have been reported, and prototype materials are now available. Low-Dielectric-Constant Materials Aluminum oxide-based ceramics and most glass-bonded ceramic materials have dielectric constants in the range of seven to nine. The need for high- speed circuitry demands materials with still lower dielectric constants. Organic materials are available with dielectric constants as low as 2, but these materials are incompatible with the high-reliability hermetic ceramic packages. For inorganic materials, the lowest known dielectric constant is that of silica glass at 3.9. However, the thermal expansion characteristics and sintering characteristics of this material are incompatible with typical processing conditions. Consequently, glass-bonded ceramic compositions are being developed that strive to provide lower dielectric constants than those available in aluminum oxide-based materials. The typical crystalline phase in these materials is cordierite, a magnesium-alumino-silicate material. Also necessary are glass-bonding agents of relatively low dielectric constants, which typically avoid the use of lead oxide. Such materials, with dielectric constants in the range of 5 to 6, are now available in prototype form. Inorganic materials with even lower dielectric constants probably await the implementation of a controlled-void technology. In this technology, the requirement is to process materials so that a large fraction of pores can be included, thus lowering the dielectric constant of the matrix phase. Incorporating hollow glass spheres or possibly applying sol gel glass technology are means of introducing controlled porosity. Closed pores are necessary to avoid any loss of hermeticity or possible inclusion of impurities that might lead to poor dielectric properties If this technology could be successfully employed using silica glass ? dielectric constants below 2 might be feasible. However, this technology will always be fraught with poor mechanical properties, since the pores are flaws that concentrate stresses and reduce the effective strength. High-Thermal-Conductivitv Materials The design of integrated circuits and hybrid integrated circuits demands that the heat generated be dis s mated through the package. Organic packaging materials or plastic packages typically have quite poor thermal dissipation
82 properties, and for this singular reason ceramic packages are often used. Of the traditional ceramic packaging materials, aluminum oxide is a reasonably good thermal conductor. Of lesser conductivity are the glass-bonded ceramics, a deficiency that limits the use of some of the low-k materials. Traditionally, when very high thermal conductivity has been needed, the material of choice has been beryllium oxide. This material has a thermal conductivity roughly equivalent to that of aluminum metal and about 10 times higher than aluminum oxide. In many respects, berylluiw~ oxide is a convenient replacement for aluminum oxide; the processing conditions, metallizat~on systems, and sintering temperatures are similar. However, beryllium oxide suffers from a major disadvantage--its toxicity. The care needed in processing even the powder has driven the price to about $50 per pound. Similarly, the caution exercised in processing this powder into ceramic packages adds additional cost to these packages, thereby leading to very expensive devices. Further limiting its use are the concerns of users in handling these packages . Two new materials with high thermal conductivity have become available within the past five years: silicon carbide and aluminum nitride. Silicon carbide was first developed in a high-thermal-conductivity form by Hitachi. This was achieved by adding small amounts of beryllium oxide (less than 1 weight percent) to pure silicon carbide and sintering by hot pressing at very high temperatures, generally in excess of 1900°C. In contrast to less pure forms of silicon carbide, which are electrically conductive, this high- purity, beryllium oxide-doped form of silicon carbide is a good electrical insulator and has a thermal conductivity approximately equal to that of beryllium oxide. The shortcomings of this material are the high processing temperatures, the small beryllium oxide content which leads to some real or imagined health hazards, a high dielectric constant (on the order of 20 or more), and the lack of an applicable technology for making multilayer metallized packages. Thus, use of this material has been limited to some special substrate applicatior~s, particularly for laser mounts. The second material, aluminum nitride, now offers greater promise than silicon carbide. If processed properly, this material has been shown to have a thermal conductivity even higher than that of beryllium oxide. Furthermore, the dielectric constant of aluminum nitride is approximately that of aluminum oxide. As with many covalently bonded ceramic materials, aluminum nitride also has the disadvantage of being very difficult to sinter, requiring temperatures in excess of 1900°C. Processing must also be done in the absence of oxygen because most of the oxides formed from these tend to degrade sharply the thermal conductivity of this material. However, some oxide dopants have been shown to be advantageous in aiding sintering without severe degradation of thermal conductivity; examples of dopant oxides are magnesium oxide or yttrium oxide. Another potential shortcoming of aluminum nitride is the difficulty in bonding it to metal. (Aluminum nitride has been traditionally used as a crucible material for melting metals because of its tendency to not bond to metals.) However, for multilayer ceramic substrate applications, bonding co metals is crucial. Nevertheless, the current state of the technology is better than that for silicon carbide since some laboratories have demonstrated processes for satisfactorily bonding metals to aluminum
83 nitride. Generally, additives to the metals or additives to the ceramic are necessary to initiate this adhesion. The material of highest known thermal conductivity is diamond, a material used in special cases as substrates and being considered for future more extensive use as substrates. Sumitomo Electric in Japan markets single crystalline diamond substrates for use in very demanding applications. These single crystalline substrates are, of course, very expensive, and means of potentially reducing the cost of such substrates are being explored. One possible fabrication technology is to sinter lower-cost diamond powders into dense polycrystalline diamond substrates; however, the state of this technology is still uncertain. The incorporation of electrical defects in the diamond structure, or having graphitic layers or other impurities present, has led to low electrical resistivity, which makes these sintered substrates unsuitable for electronic applications. A promising means of fabricating diamond for electronic applications lies in some of the CVD methods that have been explored in the mid-to late 1980s. This technology is still limited to relatively thin layers, generally less than 20 Am thick, and they also show irreproducible quality in their electrical insulating characteristics. Figure 5-2 (also Appendix E), cited earlier, reviews typical thermal conductivity and dielectric constant relationships of some important substrate materials. 20 to -5 -1 0 10 20 THERMAL CONDUCTIVITY (W/cm OK) Figure 5-2. Coefficient of thermal expansion and thermal conductivity ranges available with composite materials.
84 POLYIMIDES IN HIGH- DENS ITY PACKAGING Polyimides have a number of characteristics that make them important for high-density electronic packaging. Some of these features are discussed in the following, and some insight on their more extensive use in electronics is provided ~ Trends in High - Dens ity Packaging Until recently, computer systems with high- density packaging have used one of three approaches to chip interconnection: (a) conventional epoxy boards and multichip ceramic modules single chip carriers, (b) conventional epoxy boards and multich~p ceramic modules with single chip carriers, or (c) conventional epoxy boards with chips directly attached to multichip ceramic modules. None of these systems had the capability of wiring all chips together with a s ingle packaging level. The increasing size of central processing units (i. e., the number of devices ire a CPU) and the importance of eliminating interconnection delays have led to exploration of approaches that have the potential for eliminating significant fractions of interchip signal delays. The published literature contains numerous descriptions of efforts to reduce the gap between wiring density (and concomitant signal delay) on the chip and in the package. Two of these are (a) the silicon chip carrier, in which the interconnection between logic chips is accomplished by multilevel thin-film polyimide insulated copper or aluminum wiring on a silicon wafer, and (b) the hybrid ceramic module with levels of polyimide board, which is analogous to the traditional epoxy board, but has a lower thermal expansion coefficient and higher thermal stability for attachment of chips with a TAB- like carrier. However, only recently have these approaches to high-density packaging appeared in announced computer products. The distinction between CMOS, bipolar, BiCHOS, and GaAs logic technologies does influence the optimum approach to high-density, high- performance packaging. CMOS and BiCMOS chips tend to have a greater number of logic devices per chip compared to bipolar and GaAs. This leads both to greater interconnection complexity for bipolar and GaAs logic and to an increased emphasis on the performance of the interconnection (packaging) technologies for these chip families. Despite this distinction, the technical literature abounds with reports of thin-film polymer-copper or aluminum interconnection for all the chip technologies. Consequently, it is possible to conclude that efficient chip interconnection has become important to all chip logic technologies. Polvimide Processing The processes used to fabricate polyimide-based packaging technologies vary widely. In the extreme of "chip-like" processing, conventional chip processes of photolithography, RIE processing, and spin coating are employed. In another extreme, conventional board processes are employed where the
85 prepreg is fabricated, electroplating is used to form metal lines, and lamination, drilling, and electroplating are used to interconnect the levels. Combinations of these basic approaches make use of mixes of both extremes. One example is the use of "chip-like" processing but using "metallization" with electroplating of copper. No distinct trends in use of these processes are observable today, probably because the use of polyimides in high-density packaging has yet to reach maturity. PolYimides and Properties for High-DensitY Interconnection The polyimides currently used in the foregoing technologies are numerous, but all have a common characteristic. They are used when epoxies lack adequate properties for the application. This generally occurs when improved thermal properties (e.g., high direct solder-attach temperature) are required, improved mechanical properties are needed (polyimides generally are much less brittle and have greater tensile strength), or a lower dielectric constant is required. The thermal properties of polyimides are an improvement over epoxies in two ways: First, the thermal stability is considerably greater, and second, the glass transition temperature, or the temperature at which the polymer softens, is most often greater than 280 ° C . The dielectric constant of polyimides is usually between 3.4 and 2.9, as compared to epoxies, which are greater than 4.0. Today, many polyimides are available for high-density packaging applications. The properties of these materials range widely in a number of categories: glass transition temperatures between 250 and 400°C are available; mechanical properties ranging from brittle to very tough may be achieved; thermal stability at 250 deco 400°C is possible; dielectric constants from 2.65 to 3.4 are typical; adhesion to the ceramic, polyimide, SiO2 , or metal interfaces varies widely, but generally requires an adhesion-promotion agent; the stress in a polyimide film varies widely among commercially- available materials (amorphous polyimides generally have a film stress of about ~ to 7 Ksi, whereas ordered polyimides may have film stress as low as 1 Ksi); and planarization also varies widely between polyimides but is generally dependent on the percent of solids in the formulation. The thermally-induced stress in polyimide packaging structures is usually dominated by the high ICE of these materials. Recently, a breakthrough has occurred with the availability of low-TCE polyimides. These materials may have linear expansion coefficients of 6 X 10-6/° C compared to 35 x 10-6/°C for conventional polyimides. In addition, these materials may be tailored for TCE matching. This new family of materials will likely be used widely in high-performance thin-film interconnections. While these materials offer a significant advance in the reduction of stress in interconnection structures, further improvements in self-adhesion of these materials are desirable. Although the packaging process engineer has this wide variety of properties to choose from, it is generally true that the best properties in each category are not available in a single material. Consequently, selection
86 of a polyimide for a packaging application may be a complex process in which compromises must be made. Photosensitive Polyimides The processing of polyimide films for high-density packaging is often complicated by the necessary patterning of the films. This patterning is generally accomplished for board-like structures by drilling, by oxygen reactive ion etching (RIE) for fine-line features, or by etching with a strong base or hydrazine. Each of the last two methods requires separate masking steps that add to the complexity of processing. Considerable simplification of serial-process fabrication of polyimide-based packaging technologies can be realized with photosensitive polyimides. These materials have been commercially available for some years from a number of chemical companies in the United States, Europe, and Japan. Japanese chemical companies, in particular, have been most aggressive in making photosensitive polyimides commercially available. Again, as in the case of conventional polyimides, choosing a photosensitive polyimide for a packaging application is a complex process in which the optimum combination of properties is not available and compromises must be made. One example is the fact that most photosensitive polyimides shrink about SO percent in processing from the lithographic imaging to final cure. This shrinkage leads to limitations in feature size resolution, which currently is in the range of 25 Am for thicknesses greater than the 8 Am thickness used in packaging technologies. General Availability of Technical Information on Polyimides Packaging engineers who seek to use polyimide materials are often confronted with the need for detailed information on their chemical, physical, and process characteristics. Chemical companies have traditionally supplied customers only sketchy information on these materials. University research on the chemistry physics, and process characteristics has not been a popular endeavor, as this has been considered an "old" field of research. Large companies that intend to use polyimides in new technologies have conducted their own R&D but have not published this information widely. Consequently, detailed scientific understanding of polyimides, which forms the basis for a well - engineered packaging ~cechnology, has not been available to the whole industry. The technical vitality of the U.S. high-density packaging industry would be well served if both chemical companies and universities placed greater emphasis on providing this information. Benzocyclobutane (BCB) is a polymer recently introduced that has great promise as an interlayer diel ectric and other applications. The dielectric constant i s about 2.7 and the loss is very low, as expected for hydrocarbons . Polymerization does not release volatiles and high glass transition temperatures, greater than 350°C, are achieved. This material exhibits very low moisture absorption and is chemically extremely stable. Although there is much less industrial experience with BCBs, some commercial applications are already appearing and BCBs could become the polymer dielectric of choice in a
87 wide range of circuits. A photoimageable BOB has not appeared yet, but it can be plasma etched through a lithographic mask. Polymers in Future High-Densitv Interconnection Technologies The trend toward thin-film interconnections in high-density packaging appears to be set and will continue for the foreseeable future. With the evolution to higher-performance requirements, incremental improvements in the properties of polyimides or polymers used as insulators will be sought by the industry. Required improvements will be needed by the industry, and will probably occur in two categories: (a) photosensitive polyimides with greater lithographic image resolution and (b) polymer insulators with lower dielectric constants. It would be useful if easily processible materials with dielectric constants as low as 2 were available. These requirements present a significant challenge to the chemical industry because they are not currently available . TAPE AUTOMATED BONDING Tape automated bonding (TAB) uses a premanufactured lead frame as a substitute for wire bonding. The lead frame presents a uniform array of inner leads to be attached to the bond pads on the surface of the die. The die and leads are then excised from the lead frame, and the outer leads are bonded to a package or a multichip substrate. The TAB lead frames are manufactured in a continuous process and are available in reel, strip, or slide format. All the leads can be bonded to the die simultaneously. TAB is often preferred over wire bonding for multichip modules because it can accommodate more leads per chip (more than 300) and higher clock frequencies. There is a cost overhead for the tape design and tooling for each chip size, and, therefore, TAB is best suited to high-volume applications. TAB is routinely used to fabricate liquid crystal television displays, which require many leads for relatively small die sizes. One brand of hand-held televisions contains TAB with 80 Em lead pitch (about 300 leads per linear inch). The first U.S. applications for high-lead-count TAB will be in computer work stations. This application requires packaging of high-lead-count devices (up to SOO per chip), with operation clock rates of 50 to 100 MHz, and the ability to be manufactured in volumes of tens of thousands. Once these requirements are met, TAB use will then spread to many other applications . Wire bonding will continue to be used for single-chip packages with up to 200 leads and also for prototyping of multichip modules. From a materials point of view, TAB consists of copper conducting fingers carried on a polyimide film. The joint between the TAB tape and the chip bonding pad requires a metal "bump," either placed on the chip or on the tape. Because TAB tapes are very thin, the mechanical modulus of the polyimide is an important factor in simultaneously positioning all of the lead fingers in register with the bonding pads. Thermal expansion is also important.
88 The expendability of TAB technology from the present 200 I/Os per chip to the expected 600 I/O chips of the mid- 1990s will require much finer-pitch inner-lead peripheral connections or area array chip - to - TAB connections . Both approaches are difficult, if not impossible, with present TAB substrate materials. TAB carriers with lower TCE and higher modulus are required. The first of these improved substrates is available in the form of a lower-TCE polyimide, but further substrate materials improvements will be required to reach the mid-1990s requirements. TAB has been a promising technology for many years, but it is still not widely used in the semiconductor electronics industry. At this time, nearly every major semiconductor manufacturer is either using or evaluating the technology. The committee was not in complete agreement regarding the degree of market penetration and the time scale for penetration. DIAMOND Diamond is among the most interesting of all materials (DeVries, 1987). In the familiar single-crystal form, it is far too expensive to be employed in ordinary packaging * The phys ical properties are remarkably attractive, however, and efforts to grow diamond, or di~mond-like, films have persisted through the years. Within the past decade, films have been successfully produced by chemical-vapor-deposition (CVD), and research activity in this technology has accelerated. Ion beam epitaxy has also been employed to produce films with good properties that can be deposited on various substrates. Although growth rates are small and conditions involve moderately high temperatures for the substrate, good progress is being made, and it is reasonable to expect that diamond-film passivation layers will become possible within a few years. Among good dielectrics, diamond has uniquely high thermal conductivity, and CVD films have approached values typical of bulk crystalline diamond. The value of 20 W/cm°K at room temperature is an order of magnitude better than BeO and four orders of magnitude better than epoxy. Hardness, relatively low dielectric constant, and good optical properties also favor diamond films. At liquid nitrogen temperature, thermal conductivity is above 100 W/cm°K. Applications that would benefit from the singular properties of diamond films include chip passivation, substrate coatings, and interlayer dielectric layers. The design possibilities are sufficiently removed from present materials so that new strategies of thermal management could arise based on the s e new diamond f i lms . SUPERCONDUCTORS It has been recognized for many years that superconducting materials offer great promise for high-speed electronics (Hunt, 1989~. A number of organizations have worked on the development of Josephson junction technology, and at least one company, Hypres, has successfully marketed commercial devices. There are important differences between silicon electronics and
89 Josephson junction electronics (e.g., voltage level), and it has been difficult to overcome the momentum of the more established technology. Not the least of the Josephson burden is the requirement for liquid helium cooling. With the discovery of high-temperature superconductors (HTSCs), the comparisons have been reopened, because liquid nitrogen cooling is easier and more economical than cooling with heli~,rn. The problem of packaging and interconnection of Josephson circuits is beyond the scope of this report. There is, however, the possibility of employing superconducting striplines as interconnects in printed circuit and multichip module structures. As lateral dimensions of circuit lines shrink (see Figure 2-8), the line resistance can become a problem; for example, a 2 Am x 8 Am copper line has a resistance of 10 ohms per cm at room temperature but only about 2 ohms per cm at liquid nitrogen temperature. As the line resistance becomes an appreciable fraction of the line impedance (50 to 70 ohms), the speed of signal transmission decreases, a feature that favors low- temperature operations. The new HTSCs are ceramic materials that will require considerable development before their promise can be realized in practical circuits. These materials must be processed at high temperatures and must be protected from atmospheric moisture. In addition, electrical contacts with them to other metals are not easily made, and the properties at high frequencies are less than ideal. These are early days in the HTSC field, but, with a massive level of research under way world-wide, the problems will be overcome. Interconnects for advanced electronics are under consideration as an early application. Unlike superconductor wire applications, much of the ductility and flexibility requirements in interconnections are less demanding. There can be little doubt that HTSCs will become a part of interconnect structures, although applications are probably several years away. Research on all aspects, scientific and technological, should be strongly encouraged, bearing in mind that materials interactions and process compatibility will be particularly important. COMPOSITES Composite materials represent the extent to which materials are being engineered to achieve design intent; for example, strong fibers in polymer matrices have had an enormous impact on the construction of aircraft and other aerospace structures. In electronics, composites have long been used as printed wiring substrates (e.g., FR-4) and silica-filled encapsulation epoxies. Today, new composites are beginning to appear that will broaden the range of materials combinations found in electronic structures. The more traditional fiberglass substrates (woven E-glass mats embedded in a bisphenol-A epoxy matrix) are being replaced by materials that give better dimensional control, lower dielectric constant, and greater operating temperature range. Polyimide fibers can be designed with small and even negative coefficients of linear expansion, and these are beginning to find their way into electronic substrate composites. Expanded teflon fibers may be
go introduced into epoxy matrices to achieve lower dielectric constant. More highly functionalized epoxy matrices offer better dimensional control and higher glass-transition temperatures. Polyimide matrices also offer high- temperature performance. This latter area is one of intense activity, and new systems can be expected to increasingly penetrate the high-density PUB market. Research in this area will benefit from closer coupling with systems designers. Filled-plastic encapsulants today employ silica almost exclusively, but the opportunity exists to enhance thermal conductivity by introducing alternative fillers. The thermal conductivity of the encapsulation compounds can be improved by factors of two to six by introducing compounds such as alumina, magnesia, and boron nitride at the expense of larger (less than 50 percent) coefficients of thermal expansion (CTE). Research in this area could yield important results (see the first section of this chapter). Ceramic-matrix and metal-matrix composites also are available, and the field is developing. Opportunities for matching CTEs of composites and semiconductors while improving thermal conductivity are illustrated in Figure 5-2 cited earlier. Their use in printed wiring and multichip modules is suggested, and some applications already exist; these materials also have been used for housings for microwave and power units. This is an area in which significant work is being done, both in the United States and overseas, but the work is not very visible. Improved composite materials for packaging and interconnection should be a significant focus of future work because the opportunity to design materials with desired substrate properties is worthy of further attention in the context of high-density electronics. The high thermal conductivity of carbon fibers suggests their use in composite materials for heat sinks. Carbon-fiber-reinforced copper, for example, can be designed for good thermal conductivity coupled with weight reduction. Also, carbon-fiber-reinforced epoxies can be made with high thermal conductivity, dimensional control, and ease of fabrication. In these , continuous fibers are preferred. Here again, composites offer the flexibility to engineer materials with desired properties. Multichip modules, as discussed in Chapter 3, offer high-density interconnections on a planar surface. These modules may have hundreds of I/Os, usually arranged around the periphery of the module. They may be attached to a mother board through pin grid or pad grid array terminals, or they may be combined in a stacking fashion to form supe`modules with short interconnect paths. The z-dimensional connections can be made by special composite materials that electrically conduct in the z-direction, yet are highly resistive in the x- and y-directions. The conductors are metallic and are oriented In a matrix that may be ceramic or polymeric. Photodefined holes are etched in the ceramic substrate, then filled with wadded-up wire ("fuzz- button"~. This arrangement can interconnect facing surface pads reliably without solder; i.e., pressure alone suffices (see Appendix D).
91 Polymer systems can be made to conduct by using magnetic alignment of nickel balls, oriented wires, or balls of metal whose diameter is greater than the layer thickness. These composites may be held in place mechanically or by adhesives. Many contacts can be made in a single operation, thus facilitating assembly. These composite anisotropic conductors are still in an introductory phase and have not been demonstrated in high-reliabili~cy, high- density electronic systems. They have been employed in small television displays, where a large number of contacts (1800) is required and disassembly is important. These techniques offer ease of assembly and other advantages, but much testing remains to demonstrate the reliability needed in high- density electronic systems. MATERIALS FOR VERY-HIGH-FREQUENCY DIGITAL SYSTEMS Very-high-frequency digital electronic sys tems are extremely demanding in terms of materials properties. Generally, inorganics are used as substrates for the first level of interconnects, whereas organics serve as interlayer dielectrics. High-frequency structures require low-dielectric- constant values, whereas the power plane will need very high (' to store charge very close to the chips it is powering. Inorganic dielectrics generally have high dielectric constants, i.e., ('2 3.8. Thus, they are not likely materials of choice for interlayer dielectrics. Ceramic layers can be prepared in porous forms that can achieve (' < 2, but it remains to be demonstrated that these materials have sufficient mechanical durability and can be prepared with a smooth surface. Process compatibility for building multilayer structures must be established and long- term stability characterized. Even so, this is an interesting class of material that offers radiation resistance and low high-frequency loss. Ferroelectric ceramics can offer high - dielectric - cons tent ~ ~ ' >103 capability for essential charge storage near active devices. For high- frequency applications, it will be necessary to demonstrate that the material can respond rapidly enough to deliver the charge in the time scale required. In additions the dielectric constant of ferroelectrics often decreases dramatically as frequency is increased through the GHz range. Polyimides have been favored as ~nterlayer dielectrics for MCMs, and a great deal of development work has been focused on such multilayer structures. As circuit speed is pushed higher, lower-dielectric-constant materials will be required, and polyimides will be displaced. Polyimides absorb water, which leads to E' variability and high-frequency losses (I"). Hydrocarbon and fluorocarbon polymers are most promising for ['<3 and low dielectric loss (~tt<10-3). Considerable development effort will be required to provide low- dielectric-constant layers that are process-compatible, exhibit good adhesion, and satisfy all the other design requirements.
92 MATERIALS FOR CONNECTOR APPLI CATIONS Connectors are necessary in high-density electronic systems to facilitate assembly, repair, and maintenance. Unfortunately, connector design historically has been handled as an afterthought to system design. Although this approach has created serious reliability problems in the past, it has not necessitated a redesign of entire systems. As system performance increases, as measured by signal rise time, electrical discontinuities that are transparent at large rise times cause serious signal reflections. Physical design of connectors that provide thousands of connections and are capable of handling signals with less than 100 psec rise times are a significant technical challenge. From a materials standpoint, advanced connectors will be driven to the most reliable metallurgy, e.g., gold mated to gold on palladium. The choice of insulator is made on the basis of mechanical strength. Because of the very small cross sections of the molded parts, thermoses materials are favored-- e.g., polyethyleneterephthalate, polybutyleneterephthalate, and polyphenylene sulfide. For higher-temperature applications, polyethersulfone and polyetherimide can be employed. Physical design of high-density connectors is becoming a major challenge, with materials properties and processing occupying an important position in meeting the performance needs. THE THERMAL CONDUCTION MODULE IBM's thermal conduction module (TCM) has been at the leading edge of integrated circuit chip packaging for some time. This very significant accomplishment has taken a long time to develop since its beginning in 1964. This concept, and early work, started in a program with the appropriate name, "next-generation technology" (NGT). A technical description of the TOM was published by Blodgett (19831. The following discussion briefly describes the his tory of how the TOM evolved . The NOT program was started in January 1964, and in April 1964, IBM announced the System 360 with its solid logic technology ~ SLT) . The SLT program made use of small silicon chips containing single transistors that were attached to 1/2- by 1/2-in. ceramic substrates containing 16 pins that were soldered into printed-circuit daughter cards. Small copper spheres, plus solder, were used to connect the inverted transistor chips to the thick-film pattern on the ceramic substrate. Resistors were fabricated on the chip by a silk-screen process, as was the wiring on the chip. The resistors were later trimmed to value by a "sand-blasting" operation. Within IBM, in 1964, there was a small, but articulate, faction that argued that SLT was threatened by integrated circuits. This issue of SLT versus integrated circuits quickly became polarized and highly emotional. The NOT program, from its inception had, decided that integrated circuits and monolithic memories were the correct technical thrust for the future, and an effort was made to stay out of the highly emotional issue, but with little success. How to package integrated circuits was not resolved within the NOT
93 program until about August of 1964. At that time, the NGT program defined a four-phase approach to solve the packaging problem of integrated circuits. For phase 1, it was proposed to put integrated circuit chips on the SLT substrate, which would provide the benefits of integrated circuits without having to develop a whole new package. In addition, the SLT module was consistent with the level of integration on chips at that time. Phase 1 later became known as the monolythic systems technology (MST), which was the technology used in IBM's System 370. NOT's phase 2 proposed using a 1 by 1 in. ceramic substrate, which, with its larger area and more I/O pins, could support chips with higher levels of integration. Phases 1 and 2 were based on established packaging technologies, while still concentrating on learning how to produce integrated circuit chips to go on those packages. Phases 3 and 4, being further away in time, proposed a significant departure from the SLT module. The NGT program recognized that the level of integration would constantly increase with time, and thus a viable packaging strategy must take this into account. It also was recognized that the exponent in Rent's rule was less than 1. This meant that the ratio of I/O pins to circuits would decrease with increasing levels of integration on the chip, a reasoning that held for the next level of packaging as well. Contacts between levels of packaging take up space, are costly, limit performance for high-performance systems, and can create reliability problems. By going to higher levels of integration at all levels of packaging, one or more levels might be eliminated. During the period from 1964 to 1966, the NGT program had a joint effort with Texas Instruments (TI) that was structured to explore the future problems of large-scale integration (LSI). In this joint program, TI supplied IBM with wafer-scale integration consisting of 9 wafers with 120 circuits per wafer mounted in 9 modules. A small system, using these modules, was working by early 1966. TI used pattern interconnection in fabricating the wafers. At that time, fixed pattern interconnection could not provide LSI capability. The NGT program selected multilayer ceramics (MLC) as the technology with which to construct the modules for phases 3 and 4. Phase 3 was targeted at future cost-performance types of systems, whereas phase 4 was focused on future high-performance systems. The plan in phase 4 was to use high- performance emitter coupled logic (ECL) on integrated circuit chips. High- performance circuits consume high power per circuit. However, since one NGT objective was to place as many circuits on the chip as the manufacturing process would allow at a given time, a way had to be found to dissipate all the power produced by the many chips on a module. It was felt that fluorocarbons could be used in the module to carry heat from the chip to a heat exchanger on the module. Use of an MOM with signal conductors embedded in ceramic that had a dielectric constant of approximately 9 caused a debate within IBM. The debate ensued over the advantages of transmitting a signal from one chip to another
94 through a relatively low-dielectric-constant material, such as FR-4 epoxy, versus a high-dielectric-constant material, such as alumina ceramic. The velocity of propagation is higher in FR-4 than in ceramic, but the distance can be much shorter in ceramic. Only relatively recently has a consensus emerged that supports the NGT researchers. By early 1966, IBM decided that a replacement was needed for System 360. To do this, a monolithic systems technology a (MST) program was initiated based on the NGT phase 1 proposal and staffed by NGT people who had been previously asked to concentrate on the highly aggressive phase 4 program. At this time, the NGT program had rudimentary multichip, multilayer ceramic module hardware working. Domestic work was stopped at IBM. Fortunately, the NGT program had generated interest in multilayer ceramics in IBM's Boeblingen laboratory, and researchers there continued and improved on the multilayer ceramics work that had been started in IBM's domestic laboratories. The Boeblingen effort became the basis for IBM's present TOM when this technology was reintroduced to the United States in the mid-1970s. The purpose of recounting the early history of multilayer ceramics at IBM is to emphasize the point that developing new structures employing new materials is a long and difficult process. When such a process is successful, the payoff can be quite dramatic, as has been demonstrated at IBM. REFERENCES DeVries, R. C. 1987. Synthesis of Diamond Under Metastable Conditions. Annual Reviews of Materials Science, vol. 17, pp. 161-187. Hunt, V. Daniel. 1988. Superconductivity Sourcebook. New York: John Wiley ~ Sons' pp. 38-39. Blodgett, A. J. 1983. Microelectronic packaging. Scientific American, vol. 249, pp. 86-96, July.
